OpenAI Stretches Its Price Ladder From $2 to $500

What Happened

On September 29, OpenAI held DevDay 2026 in San Francisco and spent a 53-minute keynote unveiling more than twenty products, prices and plans. The headline item was Dots, a persistent agent that keeps working inside ChatGPT after the user logs off. According to Techloy's rundown, one Dot is included with Pro and Business Premium plans, which start at $100 a month, and the feature is unavailable in most of Europe and the UK at launch.

The model news came with a twist. GPT-6.1 Sol is a budget model priced at $2 per million input tokens and $10 per million output tokens, roughly one-fifth of what GPT-6 Astra charges at its standard rate. It carries a 1.05-million-token context window, five reasoning-effort levels, and a cached-input price of $0.10 per million tokens. Sol is available to ChatGPT Plus subscribers and above. Reporting around the event also noted that a more powerful GPT-6.1 Astra had been pulled the day before the keynote over safety concerns, and the presentation never mentioned it.

At the premium end, OpenAI introduced an Ultrafast speed tier that generates up to 300 tokens per second, about eight times the standard pace in Codex and six times in the API. On GPT-6 Astra, that speed is priced at $60 per million input tokens and $300 per million output tokens, six times the standard rate. A new Pro 500 plan sits on top of the consumer lineup: $500 a month for 25 times the Plus allowance, with Ultrafast access and no five-hour limit.

The plan changes cut the other way for existing customers. Starting October 30, the $200 Pro tier drops from 20 times to 10 times the Plus allowance for Codex and ChatGPT Work tasks, and weekly GPT-6 Pro messages fall from 200 to 100, per Digital Today. Current subscribers keep their quotas for a transition period and receive one-time credits.

Why It Matters

Taken together, the announcements describe a price ladder far wider than anything OpenAI has offered before. At the bottom, Sol gives developers a model at one-fifth of the flagship's cost. At the top, a developer or power user can spend $500 a month and pay six times the standard token rate for speed. The analysis at Capital & Compute put it bluntly: the cheap tier keeps developers on the platform, while the expensive tier is where OpenAI expects to earn.

This is also a distribution story. ChatGPT is reported to reach around 1.2 billion weekly users, and the launch leaned on that base. A feature called Sign in with ChatGPT lets users spend their ChatGPT allowance inside third-party apps, with 16 launch partners including Notion, Vercel and Devin. A separate Marketplace with more than 30 partners lets enterprise customers spend their OpenAI commitments on partner products. Both aim to turn a subscription into a currency that works outside OpenAI's own apps.

For developers, the pricing math is the practical takeaway. A cached-input rate of $0.10 per million tokens makes long, repetitive agent workflows much cheaper on Sol, while latency-sensitive coding loops can pay a premium on Astra. Choosing between the two is now a product decision rather than a simple upgrade path, and teams that standardize on one model risk either overpaying or leaving speed on the table.

Reaction

Early reaction was mixed. A founder-focused review credited the event with more than twenty launches but flagged friction. Early tester Dan Shipper found Dots useful yet said it was "too buggy for me to recommend now," and the review described permission problems and dropped messages in testing. The live demo also had awkward moments.

The sharper criticism targeted the plan changes and the distribution promises. The same review argued that OpenAI has tried similar ecosystem plays before, with the GPT Store and Instant Checkout producing variable results, and suggested treating the new partner channels as a cheap way to be available rather than a dependable growth engine. It also pointed out the cost of the Pro reshuffle: a plan that a workflow was built around can lose half its Codex allowance overnight. For teams budgeting around a fixed monthly subscription, that is a reminder that usage tiers are a moving target.

Sam Altman, by contrast, was openly enthusiastic about Dots, saying he had been waiting for such a product for years. Whether customers share that enthusiasm will depend on how quickly the early reliability problems are fixed.

What's Next

Several pieces are still unfinished. Pricing for the Decisions API, which is in a limited preview, has not been announced. Ultrafast for Sol is described as coming within days, also without a price. Dots will need a pricing model beyond the single included agent, and a Private Inference rollout is expected this fall. The coverage also points to a consumer hardware device expected later in 2026, which would extend ChatGPT beyond the phone and laptop.

The cross-device continuity feature announced at DevDay, which lets a coding task start on mobile and finish on desktop, is an early hint at that direction. The near-term test is October 30, when the Pro changes take effect and the first real reaction from paying customers arrives. Watch for how many $200 subscribers move up to the $500 plan, how many move down, and whether Sol's low price pulls developer spend away from Astra.

Closing Thoughts

It is tempting to read DevDay as a feature parade, but the more interesting story is the pricing architecture underneath. OpenAI is segmenting its audience more finely than before, with a cheap model for volume, an expensive speed tier for urgency, and a subscription ladder that nudges heavy users upward. That is a normal playbook for a maturing software business, and it signals that the company is thinking harder about margins and not just reach.

The risk is trust. Shelving a model the day before a keynote, trimming a paid plan's allowance in half, and shipping an agent that still drops messages all invite scrutiny at the moment OpenAI is asking customers to rely on it for more. The company is widening the ladder, and the next few weeks will show whether customers climb it or look for the exits.

한글 요약

OpenAI는 9월 29일 샌프란시스코에서 열린 DevDay 2026에서 20여 가지 발표를 쏟아냈습니다. 로그아웃 후에도 일하는 상시 에이전트 '닷츠(Dots)', 입력 100만 토큰당 2달러·출력 10달러로 GPT-6 Astra의 약 5분의 1 가격인 GPT-6.1 Sol, 초당 최대 300토큰의 Ultrafast 속도 옵션, 월 500달러의 Pro 500 요금제가 핵심이었습니다.

이번 발표는 저가 모델부터 고가 속도 옵션과 구독 요금제까지 가격 사다리를 크게 넓힌 것이 특징입니다. 개발자는 Sol로 비용을 줄이고, 속도가 중요한 작업에는 Astra를 선택하게 됩니다. 반면 기존 월 200달러 Pro 요금제는 10월 30일부터 코덱스·ChatGPT Work 사용량이 20배에서 10배로, GPT-6 Pro 주간 메시지가 200개에서 100개로 줄어듭니다.

초기 반응은 엇갈립니다. 닷츠는 유용하지만 아직 불안정하다는 평가가 나왔고, 파트너 유통 전략이 실제 성장으로 이어질지에 대한 회의론도 있습니다. 앞으로는 Decisions API 가격, Sol의 Ultrafast 출시, 하반기 하드웨어 기기 소식과 함께 10월 30일 요금제 변경 이후의 고객 반응이 관전 포인트입니다. 참고: Techloy, Capital & Compute, Digital Today, Superframeworks.